Is there evidence for pollster “herding” in current polling? Tories are due for a large victory in Scotland if local election success translates into general election votes. A new category of voter, the “Re-Leaver”, is defined. Why polls are imperfect predictors of election results.
In FiveThirtyEight, Nate Silver provides a fascinating analysis of recent US, French and UK votes that suggests polling companies might be contaminating their polls by applying conventional wisdom incorrectly.
Before Brexit, pundits analysing polls were predicting a win for Remain, yet deeper analysis suggests that the race was far more uncertain. In the French election the opposite happened, with polls predicting a tight race only for Emmanuel Macron to win by a wide margin.
Silver thinks there is enough evidence for pollster “herding” - they avoid publishing results that differ too much from what other polls show. As a result they are influenced by conventional wisdom which results in less accurate polls.
If you’ve been following our series of explainer articles on how polling works, you might wonder how there’s room for subjectivity in reporting poll results. How might this herding manifest itself in the polls we see today? Silver’s piece details how pollsters have many choices to make about things such as which turnout model to use, how to conduct demographic weighting, and what to do with undecided voters. These tweaks all add up and can make a big difference to the poll results that are finally released.
Rise of the Re-Leavers
The FT continues an excellent run of articles about the general election with an analysis of the Tory vote in Scotland and a story about the emergence of a category of voters known as “re-leavers”.
In Scotland, the recent local election results potentially indicates an increased Tory vote at the general election. When local council areas are mapped against Westminster constituencies, the Scottish Conservatives were first in 18 of Scotland’s 59 Westminster constituencies.
However, the article goes on to say that local elections are not necessarily a strong prediction of the general election result. There are differences in voting patterns between local and general elections, including the large share – a little over 10% – of local council votes won by independent candidates. The 47% turnout rate was also high by local election standards.
The FT’s analysis of the electorate through the lens of Brexit voters included a new category: “Re-Leavers”. These are people who voted for Remain last summer but now think that the government’s duty is to adhere to the plebiscite and leave the EU.
Latest polling shows the vote split with Hard Leavers (those who want out of the EU) on 45%; Hard Remainers (those who still would prefer to stop Brexit) on 22%; and the Re-Leavers on 23%.
The emergence of Re-Leavers means that parties should not think about two pools of voters split down the middle. It also means that the the Conservatives are targeting up to 68% of voters, which may explain their dominance in the polls.
Polling produces imperfect predictions
On Medium, Chris Hanretty continues his series of informative articles around polling. In a post earlier this week he modelled the election outcome based on the results of the final day of polling, then followed this up with a post on how to model the outcome from polls on any given day of the campaign.
Both articles provide a good case study for applying statistical modelling techniques to polling data. Rather than using polls as a perfect predictor of election results, the smarter approach is to instead consider polling data as more of an imperfect guide.